I agree this could actually be a way to get some decent distribution, and it is something I've been thinking a lot about this year. In this space, Lijit is the biggest (http://duckduckgo.com/?q=lijit
), and they've done decently well, at least in terms of traffic going through the system.
The big issue is that I'm not deep crawling most sites right now, and so the site: functionality is really relying on external APIs to be up to date. It feels misleading to me to offer it through that system and in any case it won't guarantee to index everything.
The alternative is of course to extend the current tech to deep crawl and index selective sites with the widget installed. That is certainly possible, but it would require a decent amount of work. So I'm torn because there are a lot of pressing features to do on the main engine.